Rapporto di Ricerca CS Simonetta Balsamo, Andrea Marin. On representing multiclass M/M/k queue by Generalized Stochastic Petri Net
|
|
- Jason Mason
- 5 years ago
- Views:
Transcription
1 UNIVERSITÀ CA FOSCARI DI VENEZIA Dpartmento d Informatca Techncal Report Seres n Computer Scence Rapporto d Rcerca CS Febbrao 2007 Smonetta Balsamo, Andrea Marn On representng multclass M/M/k queue by Generalzed Stochastc Petr Net Dpartmento d Informatca, Unverstà Ca Foscar d Veneza Va Torno 55, 3072 Mestre Veneza, Italy
2 On representng multclass M/M/k queue by Generalzed Stochastc Petr Net Smonetta Balsamo, Andrea Marn Dpartmento d Informatca Unverstà Ca Foscar d Veneza Va Torno 55, 3072 Veneza Mestre, Italy {balsamo,marn}@ds.unve.t Abstract. In ths paper we study the relatons between mult-class BCMP-lke servce statons and generalzed stochastc Petr nets (GSPN). Representng queung dscplne wth GSPN models s not easy. We focus on representng mult-class queung systems wth dfferent queung dscplnes by defnng approprate fnte GSPN models. Note that queung dscplne n general affect performance measures n mult-class systems. For example, BCMP-lke servce centers wth Frst Come Frst Served (FCFS) and wth Last Come Frst Served wth Preemptve Resume (LCFSPr) have a (dfferent) product-form soluton under dfferent hypotheses. We defne structurally fnte GSPNs equvalent to the multclass M/M/k queung system wth FCFS, LCFSPR, Processor Sharng (PS) and Infnte Servers (IS). Equvalence holds n terms of steady state probablty functon and average performance measure. The man dea s to defne a fnte GSPN model that smulates the behavor of a gven queue dscplne wth some approprate random choce. Moreover, we prove that the combnaton of the ntroduced equvalent models has a closed-form steady state probablty by the M = M property. We consder queung systems wth both a sngle server wth load dependent servce rate, and multple servers wth constant servce rate. Introducton Queung theory and (Generalzed) Stochastc Petr Nets are mportant classes of stochastc models used to evaluate system performances. Queung systems have been wdely appled to represent resource contenton systems where a set of customers competes for resource usage. However, basc queung systems cannot model the synchronzaton between concurrent actvtes. Stochastc Petr nets (SPN) can be naturally used to represent systems wth synchronzaton and concurrency and to perform both qualtatve and quanttatve analyss. An mportant problem n system performance evaluaton based on performance models s the effcency of the soluton algorthms,.e. the ablty to defne classes of models that can be analyzed by methods wth a lmted space and tme computatonal complexty. Many results have been proposed n lterature whch gve effcent solutons of some types of stochastc models under certan condtons.
3 . INTRODUCTION Sngle queung system models have been wdely analyzed by consderng varous arrval dstrbutons, servce tme dstrbutons, classes of users and schedulng dscplnes. The sngle server queue s analyzed [4, [0, [3, [9 and several results have been derved for specal cases. Queung networks (QN) extend and combne varous queung systems to represent more complex systems. QN models can be represented by defnng an assocated stochastc process that, under some exponental and ndependence assumptons, s a contnuous-tme Markov Chan (CTMC) process. Although the statonary state probablty soluton of the assocated Markov Process, under stablty condtons, can be easly defned, the computaton can soon become unfeasble due to the hgh computatonal complexty. However, under certan assumptons QNs can be effcently analyzed by applyng the product-form theorem [5, whch defnes the steady state probablty functon as product of functons of each sngle servce center state. Product-form QNs can be analyzed by effcent algorthms (e.g. [8, [8, [7, [6) that yeld a low polynomal computatonal cost. SPNs, whch are defned n terms of a set of places and a set of exponentally tmed transtons connected by arcs, and a markng, whch s the state of the net, are represented by a CTMC process whose state space s the set of all possble markngs of the net. Specfcally the computaton of performance measures s neffcent because t requres to calculate the reachablty set, whch depends on the ntal markng and whose sze grows exponentally wth the number of places of the net and the number of tokens n the ntal markng. Henderson et al. ntroduced the dea of product-form for SPN n [2 and [ smplfyng the computaton of the performance ndexes. However, the algorthms for product-form SPNs stll requre to check condtons on the reachablty set. The class of Generalzed Stochastc Petr Nets (GSPN) allows modellng nets wth both exponentally tmed and mmedate transtons ntroducng more flexblty. The underlyng process of a GSPN can be defned as a sem-markov process. Product-form for GSPN has been studed by Balbo et al. n [3 and t s based on technques to reduce the problem to the product-form SPN theorem. Investgatng the relatons between classes of queung models and GSPN models s an nterestng problem and t has been consdered by some recent research lterature ([2, [20, [4). Most of these works focus on the relatons between QNs and SPNs or GSPNs and derve some equvalence results. However a lttle attenton has been devoted to the problem of representng varous types of schedulng dscplnes of queung systems by GSPNs. To the best of our knowledge, representng schedulng dscplnes n multclass models wth fnte GSPNs s stll an open problem. In [20 the authors ntroduce a comparson between QN models and SPN models based on the representaton of multclass features by colored Petr nets. However the dfferences between dfferent schedulng dscplnes are not analyzed. Balbo et al. n [2 combne GSPN and product-form QN by replacng subsystem n a low-level model wth ther flow equvalents models. Stll lttle attenton s devoted to schedulng dscplnes. In [4 the authors observe how they can map each servce staton of a BCMP QN to a complex GSPN. The GSPN model depends on the schedulng dscplnes but t has an nfnte 2
4 . INTRODUCTION number of places and transtons for the FCFS and LCFSPR statons. Then they gve a fnte and remarkably compact representaton by a GSPN equvalent to the detaled model. The compact representaton holds the product-form condtons for GSPN showed n [3 but t does not dstngush dfferent queung dscplnes by mappng everythng n the PS dscplne. Thus t s not possble to defne on the GSPN equvalent condtons to the ones dependng on the servce center schedulng dscplne of the BCMP theorem. On the other hand the detaled representatons of queung dscplne yeld non product-form GSPN models equvalent to BCMP QN. In ths paper we present an equvalence result between two types of stochastc models. We propose a fnte GSPN representaton of a set of queung systems wth varous schedulng dscplnes. Accordng to the BCMP-type servce centers we analyze Frst Come Frst Served (FCFS), Last Come Frst Served wth preemptve resume (LCFSPR), Processor Sharng (PS) and Infnte Servers (IS) schedulng dscplnes. The man dea behnd these results s a probablstc model of the queue,.e., all the customers of the same class wat n the same place and when a server becomes free the customer whch gets the servce s chosen n a probablstc way smlarly to what happens wth the random queung dscplne. In the LCFSPR dscplne, we also choose probablstcally the customer that looses the server when a new customer arrves to the system. The advantage of havng a fnte representaton whch s dfferent for the varous schedulng dscplnes s twofold: frst t makes the analyss easer. Second t does not requre the defnton of new semantc for the GSPN accordng to the queung dscplnes. Thus exstng analyss or smulaton tools can be used wth the GSPN nets defned n ths work. The proposed results are nterestng because they allow the representaton of an M/M/k queue wth varous queung dscplnes by a compact GSPN, whch s equvalent to the queung system n term of steady state queue length dstrbuton. A practcal consequence can be that t can extend a GSPN smulator or analyzer for analyzng multclass queue systems. The only requrement s that the tool s able to model state-dependent frng rates for tmed transtons and state-dependent weghts for mmedate transtons. There s no need to support the colored model extenson to represent dfferent classes. One could also ntegrate a GSPN analyzer by a functonalty that dentfes net structures equvalent to dfferent schedulng dscplnes M/M/k queung systems and then t can apply the closed form steady state formula. For the FCFS and PS dscplnes, the GSPN structure complexty,.e. the number of places and transctons, grows lnearly just wth the number R of classes of users, for the LCFSPR t grows lke O(R 2 ). We gve a GSPN model for the queung dscplnes consdered n the BCMP theorem [5. An open problem and a possble further research s the analyss of the GSPN models obtaned by combnng these blocks. Possbly, under some assumptons, t s possble to defne equvalence between the GSPN steady state probablty functon and the BCMP queung network steady state probablty functon. 3
5 2. GENERALIZED STOCHASTIC PETRI NETS The paper s structured as follows. Secton 2 brefly revews the GSPN models recallng formalsm we chose, Secton 3 revews some results of the queung systems theory used later n the paper. In Sectons 4, 5 we ntroduce the GSPNs respectvely equvalent to the FCFS and LCFS multclass M/M/k queue. Secton 6 dscuss the GSPN models for both PS schedulng and IS systems. The proof of some theorems are gven n appendx. Fnally, Secton 8 provdes some concludng remarks. 2 Generalzed Stochastc Petr Nets In ths secton we brefly recall the Generalzed Stochastc Petr Nets (GSPN). We consder the notaton for GSPN ntroduced n [5. In order to allow markng dependent probabltes for solvng conflcts among mmedate transtons we use the technques dscussed n [9. Then n the next secton we shall present GSPN models equvalent to queung systems under varous assumptons. Let us defne a marked Stochastc Petr Net whch conssts of a 8-tuple as follows: where: GSP N = (P, T, I(, ), O(, ), H(, ), Π( ), w(, ), m 0 ) P = {P,..., P M } s the set of M places, T = {t,..., t N } s the set of N transtons (both mmedate and tmed), I(t, p j ) : T P N s the nput functon, N, j M, O(t, p j ) : T P N s the output functon, N, j M, H(t, p j ) : T P N s the nhbton functon, N, j M, Π(t ) : T N s a functon that specfes the prorty of transton t, N, m N M denotes a markng or state of the net, where m represents the number of tokens n place P, N, w(t, m) : T N M R s the functon whch specfes for each tmed transton t and each markng m a state dependent frng rate, and for mmedate transtons a state dependent weght, m 0 N M represents the ntal state of the GSPN,.e. the number of tokens n each place at the ntal state. We consder ordnary nets,.e., functons I, O and H take values n {0, }. For each transton t let us defne the nput vector I(t ), the output vector O(t ) and the nhbton vector H(t ) as follows: I(t ) = (,..., M ) where j = I(t, P j ), O(t ) = (o,..., o M ) where o j = O(t, P j ) and H(t ) = (h,..., h M ) where h j = H(t, P j ). Functon Π(t ) assocates a prorty to transton t. If Π(t ) = 0 then t s a tmed transton,.e., t fres after an exponentally dstrbuted frng tme wth mean /w(t, m), where m s the markng of the net. If Π(t ) > 0 then t s an mmedate transton and ts frng tme s zero. We say that transton t a s enabled by markng m f m I(t a, p ) and m < H(t a, p ) for each =,..., M and no other transton of hgher prorty s enabled. We consder just 4
6 2. GENERALIZED STOCHASTIC PETRI NETS two prorty levels, 0 and. Hence when an mmedate transton s enabled all the tmed ones are dsabled. The frng of transton t changes the state of the net from m to m I(t )+O(t ). The reachablty set RS(m 0 ) of the net s defned as the set of all markngs that can be reached n zero or more frngs from m 0. We say that markng m s tangble f t enables only tmed transtons and t s vanshng otherwse. For a vanshng markng m let T α be the set of enabled mmedate transtons. Then the frng probablty for any transton t T α and any state m s denoted by p(t, m) and t s defned as follows: p(t, m) = w(t, m) t j T α w(t j, m). () Gven a tangble markng m the transton wth the lowest assocated stochastc tme fres. A GSPN s represented by a graph wth the followng conventons: tmed transtons are whte flled boxes, mmedate transtons are black flled boxes, places are crcles, f I(t, p j ) > 0 we draw an arrow from p j to t labelled wth I(t, p j ), f O(t, p j ) > 0 we draw an arrow from t to p j labelled wth O(t, p j ), f H(t, p j ) > 0 we draw an crcle endng lne from p j to t labelled wth the value of H(t, p j ), the markng m s represented by a set of m j flled crcles representng the tokens n place p j for each j =,..., M. For ordnary nets we do not use labels for the arrows. GSPN analyss conssts n fndng the steady state probablty for each tangble markng of the reachablty set. Some analyss technques are presented n [5. Under general assumptons, the stochastc process generated by the dynamc behavor of a standard SPN s a CTMC process. Mean state sojourn tmes are computed from the mean transton delays of the net. For GSPNs the dstrbuton of the sojourn tme n any markng can be expressed as a negatve exponental and determnstcally zero dstrbutons for tangble and vanshng markngs, respectvely. Thus the markng process can be studed as a sem-markov random process. The GSPN models ntroduced n ths paper present markng processes whch allow us to easly reduce the sem-markov process to a CTMC. In fact whenever a vanshng markng s reached, the next markng s tangble. Thus we can smply obtan a CTMC whose states are the tangble states of the orgnal process and the transton rates are computed weghtng the transtons rates of the orgnal process wth the frng probabltes of the mmedate transtons. Hence the mean sojourn tmes n the tangble states of the orgnal sem-markov process and the mean sojourn tmes of the CTMC are the same. Fnally let us ntroduce some other notatons: let e be an M-dmensonal vector wth all zero components but the -th whch s. We use the lower case t to name mmedate transtons, the upper case T to name tmed transtons, t to name a generc tmed or mmedate transton. 5
7 3. SINGLE QUEUING SYSTEMS WITH DIFFERENT CLASSES OF CUSTOMERS 3 Sngle Queung systems wth dfferent classes of customers In ths secton we brefly recall sngle queung systems wth dfferent classes of customers classfyng them on the number of servers and schedulng dscplnes. Let us consder an open queung system wth external arrvals, a queue, a set of dentcal servers and a set of R customer classes. The queung system s shown n Fgure. Customers of class r arrve at the system accordng to a Posson process wth rate λ r and requre an exponentally dstrbuted random servce tme wth parameter µ r, r =,..., R. The system has a set of ndependent servers, possbly nfnte. For sngle class queung systems some results n terms of steady state probablty hold for any schedulng dscplne that s work-conservatve and ndependent from the servce tme [0, [4. These results can be extended to multclass queung systems although they depend on the schedulng dscplne. We consder the followng dscplnes: Frst Come Frst Server (FCFS), Last Come Frst Server wth Preemptve Resume (LCFSPR), Processor Sharng (PS). The steady Server CLASS CLASS 2 Queue Server 2 CLASS R Server k Fg.. An M/M/k multclass queung system. state probablty of a M/M/k multclass system wth a specfc queung dscplne and constant servce rate s equvalent to the steady state probablty of a M/M/ multclass system wth the same queung dscplne and an approprate load-dependent servce rate. If all the customer servce tmes are dentcal,.e., µ r = µ for r =,..., R, the load dependent servce rate µ(j), where j s the number of customers at the system, s defned as follows: µ(j) = { jµ f j k kµ f j > k (2) 6
8 3. SINGLE QUEUING SYSTEMS WITH DIFFERENT CLASSES OF CUSTOMERS If the stablty condton R λ r kµ < holds, then we can evaluate the statonary queue length dstrbuton of the multclass M/M/k system for any schedulng dscplne by the correspondng M/M/ load dependent system. Let π (n) denote the steady state probablty of the M/M/k system, wth n = (n... n R ),.e., the probablty of fndng n customers of class for =,..., R n the system. Then we can wrte: π (n) = π 0 R λ n ( P r n R )! n R n! µ(j), (3) where π 0 s the probablty of fndng the system empty. When mean servce rates for dfferent customer classes are not dentcal,.e., µ µ j for j for some couple, j, then the load dependent servce rate functon µ r (n), for any class r and state n, s defned as follows: µ r (n) = n r n mn(n, k)µ r, n = R n. (4) The followng steady state probablty holds for LCFSPR and PS queung dscplnes: P R R π (n) = π 0 n )! R R ( R n λ n ) n j n! µ mn(k, ). (5) The stablty condton s k : n > k[ R r= λ r µ r (n) <. The BCMP theorem [5 consders servce centers wth sngle servers and state dependent servce tme. For FCFS servce statons the servce tme can depend only on the total number of customers n the system. Let n = R r= n r and x(n) be an arbtrary postve functon of n, representng the servce rate when there are n customers at the servce center relatve to the servce rate when n =. Then the steady state probablty functon s: π (n) = π 0 n! R R n! ( λ n n n µ) x(). (6) For LCFSPR and PS systems, BCMP theorem consders another state dependent servce rate. Let y r (n r ) be an arbtrary postve functon of n r whch denotes the servce rate of class r customers at servce center relatve to the servce rate when there s one class r customer at servce center.e. µ r. Then the steady state probablty functon s: π (n) = π 0 n! R R n! λ n R r= [( ) nr n r. (7) µ r y r (a) Note that these varous forms of state dependent servce rates can be combned. For example the steady state probablty (5) can be obtaned combnng equatons (6) and (7) by settng x(n) = mn(n,k) n and y r (n r ) = n r. 7 a=
9 4. REPRESENTING M/M/K/FCFS QUEUE BY GSPN 4 Representng M/M/k/FCFS queue by GSPN In ths secton we defne a GSPN that represents an R-multclass M/M/k/FCFS queue. Then we prove that the GSPN model s equvalent to the queung system n terms of the steady state probablty. Gven the M/M/k/FCFS models defned as n Secton 3 let us defne the model called GSPN-. Defnton (GSPN-). Accordng to GSPN defnton gven n Secton 2: P = P q P s {P 2R+ } wth P q = {P,..., P R } and P s = {P R+,..., P 2R }, T = T w T q where T q = {t,..., t R } and T w = {T R+,..., T 2R }, functon Π defned as follows: Π( t ) = { 0 f R + 2R f R, nput and output vectors for transton t, R: I(t ) = e + e 2R+ and O(t ) = e R+. Input and output vector for transton T R+ : I(T R+ ) = e R+ and O(T R+ ) = e 2R+, H(t ) = (0,..., 0) for all t T, w(t R+, m) = m R+ µ for R and w(t, m) = m for R, m 0 = (0,..., 0, k). Tokens arrve to places P, R accordng to Posson stochastc processes. Fgure 2 llustrates the graphcal representaton of GSPN- model where t,..., t R are mmedate transtons and T R+,..., T 2R are exponental transtons. Fg. 2. Graphcal representaton of model GSPN- Let m be a vald vanshng state of the GSPN-, and let T a T q be the set of mmedate transtons enabled by m, then the probablty of frng of t T a 8
10 4. REPRESENTING M/M/K/FCFS QUEUE BY GSPN can be wrtten as: p(t, m) = p (m) = m j {j t j T a} m j (8) We shall now derve a closed form soluton for the steady state probablty of GSPN- model by consderng the set of reachable markngs m = (m,..., m 2R+ ). Ths s gven by Lemma. Then we ntroduce a state aggregaton by defnng the aggregate state n = (n,..., n R ) where n = m + m R+, R. Ths state corresponds to the number of customers of class n the queung model. Theorem provdes the closed form soluton for model GSPN- n terms of aggregated statonary probablty of state n. Fnally the GSPN- model s shown to be equvalent to the M/M/k FCFS multclass queung system n terms of statonary probablty. Lemma. Let m = (m,..., m 2R+ ) be a reachable tangble state of the GSPN-. Then f the stablty condton holds, the statonary state probablty can be wrtten as follows: π(m) = π 0 R λ m +m R+ ( 2R =R+ m )! 2R =R+ m! ( P R m 2R )! m R m! µ(j). (9) where π 0 s a normalzng constant and µ(j) s the functon defned by (2). The proof s gven n appendx A and s based on verfyng the set of the CTMC global balance equatons. Theorem. Consder model GSPN- and let n = m + m R+, R and n = (n..., n R ) be an aggregated state. Let π a (n) be the steady state probablty of n for =,..., R. Then we can wrte: π a (n) = π 0 ( R n )! R n! r λ n P R n µ() n N R. (0) Proof. In order to derve equaton (0) we prove that: π a (n) = m m +m R+ =n R π(m), () for n N R and m n the reachablty set of model GSPN-. Consder the two followng cases: case ) R n k and case 2) R n < k. Case : R n k. Consder any combnaton of j wth r and 0 j n. Then the rght-hand sde of equaton () by usng formula (9) can 9
11 4. REPRESENTING M/M/K/FCFS QUEUE BY GSPN be wrtten as follows: j +...+j R =k =π 0 R j n π(n j, n 2 j 2,..., n R j R, j,..., j R, 0) =π 0 R =π 0 R λ n j j λ nj j λ n j j P R n P R n P R n µ(j) µ(j) µ(j) j +...+j r=k j n ( R n k)! R n k!! ( R n k)! R n k!! k! ( R n k)! R j R! (n j )! j +...+j R =k j n j +...+j R =k j n R n! R (n R j )! j! where the last sum satsfes the Vandermonde convoluton, thus we can wrte: π 0 R =π 0 R λ n j j λ n j j P R n P R n µ(j) µ(j) R ( n j ), ( R n ( k)! R R n k! n )! k ( R n )! n n,! whch s formula 0. Case 2: R n < k, that corresponds to the behavor of the queung system where all the customers are beng served and n GSPN- every place P wth R s empty. Note that n = m R+, so by equaton (9) we can wrte: π(0,..., 0, m R+,..., m 2R, l) = π 0 R λ n P R n µ() ( R n )! R n,! that yelds formula (0) and ths ends the proof. Corollary. The M/M/k queung system wth FCFS dscplne, R customer classes, arrval rates λ, R, sngle server rate µ and steady state probablty π (n) s equvalent to the GSPN- n terms of steady state probablty,.e., π a (n) = π (n) for all n N R where π a (n) s the aggregated probablty of GSPN gven by formula (0) Proof. It follows mmedately from equaton (3) and Theorem. Note that GSPN- model represents the M/M/k multclass system when the servce rate s ndependent from the class of the customers n servce. Thus t can not be used to represent LCFSPR or PS schedulng dscplnes. For example 0
12 5. REPRESENTING M/M/K/LCFSPR QUEUE BY GSPN consder the system wth LCFSPR queue, sngle server, and dfferent average servce rates for each class. Then we show now a counterexample to prove that the steady state gven by queung theory does not satsfy the GBE for the GSPN. Ths mples that GSPN- catches somehow the FCFS semantc (by not allowng preempton). Example. As example, consder a LCFSPR M/M/ queue wth two classes of customers wth average servce tme /µ and /µ 2. From queung theory we can wrte the steady state probablty as follows: π (n, n 2 ) = π (0, 0)λ n λn 2 2 (n + n 2 )! n!n 2! Suppose to represent ths system by GSPN- assocatng dfferent frng rates to transtons T 3 and T 4 : w(t 3, m) = m 3 µ and w(t 4, m) = m 4 µ 2. We calculate the effectve arrval rate to reachable tangble state m = (0, m 2,, 0, 0), wth m 2 > 0. The adjacent states are m = (0, m 2,, 0, 0), m 2 = (, m 2,, 0, 0), m 3 = (, m 2, 0,, 0), thus the effectve arrval rate to state m s: [ m 2 π(m) µ 2 λ 2 + λ (m 2 + ) µ λ 2 m 2 µ m λ 2(m 2 + ) µ 2 µ 2 m 2 + = π(m) [µ 2 + λ + λ 2. The effectve leavng rate for state m s clearly π(m)(µ + λ + λ 2 ), so the GBE on state m s satsfed f µ = µ 2. Fnally note that GSPN- can as well smulate a sngle server FCFS servce staton wth an BCMP-lke load dependent servce rate. We can state the followng lemma: Lemma 2. Let m = 2R m, x(m) be an arbtrary postve functon, m 0 = e 2R+ and let the frng rate of transton T R+,..., T 2R be w(t R+r ) = x(m)µ, r R. Then f m s a tangble reachable markng the steady state probablty functon s: π(m) = π 0 R λ m+m R+ ( R m )! R m! µ n µ n2 2 ( µ) P P 2R 2R m m. x(j). (2) The proof s gven n appendx. By defnng n = m +m R+ we can aggregate the states and we can prove that the steady state probablty π a of the aggregated CTMC s dentcal to π defned by equaton (6) of BCMP theorem. The net structure complexty s lnear on R, the number of customer classes. 5 Representng M/M/k/LCFSPR queue by GSPN In ths secton we ntroduce a GSPN whch can be consdered equvalent, for steady state probablty, to a multclass M/M/k queue wth LCFS wth preemptve resume schedulng dscplne. As we consder just exponentally dstrbuted
13 5. REPRESENTING M/M/K/LCFSPR QUEUE BY GSPN servce tmes, we do not consder the problem of representng the resume. We provde a model for ths queue system whose structure s fnte and depends only on the number of classes of customers,.e., not on the number of servers. A trval soluton could be obtaned by recallng that the steady state formula for a LCFSPR queue s equal to the Processor Sharng one so that we could use the same GSPN representaton. On the other hand we want to provde a model whch semantcally smulate closer the LCFS queue. Defnton 2 (GSPN-2). Accordng to GSPN defnton gven n Secton 2: P = P q P w P a {P 3R+ } where P q = {P,..., P R } and P w = {P R+,..., P 2R } and P a = {P 2R+,..., P 3R }, T = T q T w T f T g where T q = {t,..., t R } and T w = {T R+,..., T 2R } and T f = {t 2R+,..., t 3R } and T g = {t j,, j R}, functon Π s defned as follows: { f t T Π( t) = q T f T g, 0 f t T w Let, j R. The nput and output vectors of t T q : I(t ) = e + e 3R+ and O(t ) = e R+. The nput and output vectors for T R+ T w : I(t R+ ) = e R+ and O(t R+ ) = e 3R+. The nput and output vectors for t 2R+ T f : I(t 2R+ ) = e 2R+ + e 3R+ and O(t 2R+ ) = e R+. The nput and output vectors for t j T g : I(t j ) = e 2R+ + e R+j and e j + e R+, H(t ) = (0,..., 0) for t T q T w T f and H(t j ) = e 3R+ for t j T g, for, j R let w(t R+, m) = m R+ µ, w(t, m) = m, w(t 2R+, m) = and w(t j, m) = m R+j, m 0 = (0,..., 0, k). Tokens arrve to places P 2R+, R, accordng to Posson stochastc processes. Fgure 3 shows a graphcal model for R = 2 classes LCFSPR queue where dotted lnes are ntroduce for the sake of readablty and they do not have ant partcular meanng. Note that when a token arrves to the place P 2R+ t s temporally (.e. the state s vanshng) stored n P 2R+ and we have two cases: there s at least one free server,.e. m 3R+ > 0, thus the customer goes mmedately n servce. Ths s modelled by the mmedate transton set T f all the servers are busy,.e. m 3R+ = 0, so a customer s preempted and put n queue and the new customer goes n servce. Ths s modelled by R 2 transtons, T g. The nhbtor arcs are needed to avod pre-empton when there s at least one free server. Lemma 3. Consder the sets of mmedate transtons T q, T f, T g. Any two transtons belongng to two dfferent sets cannot be smultaneously enabled. Moreover any two transtons of T f cannot be enabled smultaneously, and f t a T g s enabled then just transtons t b T g wth b R can be enabled. 2
14 5. REPRESENTING M/M/K/LCFSPR QUEUE BY GSPN Fg. 3. Graphcal representaton of model GSPN-2. The proof mmedately derves by GSPN-2 structure. As consequence to lemma 3 we can solve the conflcts on mmedate transtons wth just one smple functon. When one or more transtons of T q are enabled, the probablty of frng for the -th transton s: m p(t, m) = p (m) = R l= m. (3) l When one or more transtons of T g are enabled, the probablty of frng s: m R+j p(t j, m) = p j (m) = R l= m. (4) R+l Now we can state a man lemma for model GSPN-2 representaton: Lemma 4. Let m = (m,..., m 3R+ ) be a reachable tangble markng of GSPN- 2 model. Then f the stablty condton holds, the statonary state probablty can be wrtten as follows: π(m) = π 0 R λ m +m R+ ( R m )! R m! ( R m R+)! R m R+! R ( ) m +m R+ µ P 2R m mn(j, k). (5) where µ s the average servce rate for one customer of class when there are no other customers n the system, k s the number of servers, π 0 s a normalzng constant. 3
15 6. REPRESENTING M/M/K/PS QUEUE AND M/M/ QUEUE BY GSPN The proof s gven n appendx A. Theorem 2. Consder model GSPN-2 and let n = m + m R+, R and n = (n,..., n R ) be an aggregated state. Let π a (n) be the steady state probablty of n for =,..., R. Then we can wrte: π a (n) = π 0 ( R n )! R n! R λ n R P R ( ) n n µ mn(k, ) n N R. (6) Proof. The proof s based on the Vandermonde formula and t s smlar the one gven for Theorem. Corollary 2. The M/M/k queung system wth LCFSPR dscplne, R customer classes, arrval rates λ, sngle server rate µ for class customers and steady state probablty π (n) s equvalent to model GSPN-2 n terms of steady state probablty,.e., π a (n) = π (n) for all n N R, where π a (n) s the aggregated probablty of GSPN gven by formula (6). The normalzng constant s π 0 = π(0,..., 0, k) = π (0,..., 0, k). Proof. It follows mmedately from queung theory and Theorem 2. The net GSPN-2 can as well smulate a sngle server LCFSPR servce staton wth a BCMP-lke load dependent servce rate. We can state the followng lemma: Lemma 5. Let m r = m r + m R+r, y r (m r) an arbtrary postve functon, m 0 = e 3R+ and let the frng rate of transtons T R+,..., T 2R be w(t R+r ) = y r (m r). The f m s a reachable tangble markng, the steady state probablty functon s: π(m) = π 0 R λ m+m R+ ( R m )! R m! R r= [( ) m mr+m r +m R+r R+r µ r a=. (7) y r (a) The proof s gven n appendx. By defnng n = m + m R+ we can aggregate the states and we can prove that the steady state probablty π a of the aggregated CTMC s dentcal to probablty π defned by equaton (7). For what concern the net structure complexty, the number of places grows as O(R) and the number of transtons grows as O(R 2 ). 6 Representng M/M/k/PS queue and M/M/ queue by GSPN The processor sharng dscplne can be easly represented consderng that the k processors are shared among the users n the system. Dfferent classes of users can have dfferent average tme servces, but all modelled by exponentally dstrbuted random varables. We can thnk that the k servers are shared among the R classes n proporton to the number of customers of the classes. 4
16 6. REPRESENTING M/M/K/PS QUEUE AND M/M/ QUEUE BY GSPN Defnton 3 (GSPN-3). Let us defne the model GSPN-3 as follows: P = {P,..., P R }, T = {T,..., T R }, Π(T ) = for each T T, I(T ) = e and O(T ) = (0,..., 0) for each T T, H(T ) = (0,..., 0) for each T T, w(t, m) = m m mn(k, m) where m = R m j for each T T, m 0 = (0,..., 0). Fgure 4 shows a graphcal representaton of the GSPN-3 model. Note that ths Posson Arrvals Fg. 4. Graphcal representaton of model GSPN-3 model s equvalent to a queung system wth PS dscplne and one server wth load-dependent exponental servce tme to smulate the mult-server feature. Therefore t mmedately follows the theorem: Theorem 3. Consder model GSPN-3. Then f stablty condton holds the statonary state probablty can be wrtten as follows: π(m) = π 0 ( R m )! R m! R λ m R P R ( ) m m µ mn(k, ), where µ s the average servce rate for one customer of class when there are no other customers n the system, k s the number of servers, π 0 s a normalzng constant. Ths model s smlar to the compact model ntroduced n [4, the only dfference s that we allow a whole state dependent frng rate thus we don t need a place whose tokens represent the total number of customers n the system. Model GSPN-3 can easly represent also the IS center. It suffces to set the frng rates of each transton T as m µ, R. 5
17 7. M M PROPERTY ON THE GSPN REPRESENTATION 7 M M property on the GSPN representaton Markov mples Markov property s ntroduced and studed by Muntz [7. In that paper he shows that f a queung system wth Posson arrvals presents departures accordng to a Posson process (M M property) then a combnaton of servce centers of ths type n a queung network has a product-form soluton. As we are consderng GSPNs we wll prove that a combnaton of GSPN-, GSPN-2 and GSPN-3 models stll holds a closed-form steady state probablty by defnng approprate traffc processes over the CTMC assocated to each of the models and usng the results gven n [6 whch generalze Muntz s work. We now brefly revew Melamed s results lmted to a CTMC n steady state. Consder an ergodc CTMC wth state space Γ and a set of traffc transtons denoted by Θ,..., Θ R, where Θ Γ Γ, Θ. Let us defne K (t) as the process whch counts the number of transtons (α, β) Θ up to t. Let m = γ Γ η Θ π(η)ξ(η γ) and for each state γ Γ let m (,γ) (γ) = η Θ π(η)ξ(η γ) where Θ (,γ) (, γ) = {β (β, γ) Θ } and ξ(η γ) s the transton rate between states η and γ. Then we can state that K (t) are mutually ndependent Posson processes f and only f the followng equaton holds: γ Γ, R R m (γ) = π(γ) m (8) We am to study the departure traffc processes from our models. Take for example model GSPN-, we can defne R traffc processes as follows: Θ = {(m, m) : m = m + }, =,..., R, (9) where m = m + m R+. In our case, n order to prove that K (t) are ndependent Posson processes when there are Posson arrvals, t suffces to prove that: γ Γ, π(η)ξ(η γ) = λ π(γ), (20) η Θ (,γ) In appendx we prove that ths condton holds for GSPN-, GSPN-2 and GSPN- 3 models by defnng approprate traffc processes. As observed n [6 ths property of the CTMC s equvalent to the M = M gven by Muntz thus t assures that a BCMP-lke composton of these GSPN models holds a closed-form steady state probablty functon. Random swtches between the blocks and user class swtches can be easly modelled by mmedate transtons. 8 Fnal remarks In ths paper we have shown how to represent mult-class sngle queung systems by structurally fnte GSPN for varous queung dscplnes. For each of the BCMP center types we have ntroduced a GSPN model whose steady state 6
18 A. APPENDIX probablty, aggregatng on the number of customers n the system for each class, s equal to the correspondent sngle queue servce center. Hence the two models are equvalent n terms of steady state dstrbuton and average performance ndexes. The man advantages of our representaton are the followng. We defne a fnte GSPN model. The abstracton level of the GSPN model allows the representaton of the queung behavor wthout ntroducng a hgh level of detals n the state specfcaton. We dstngush the customers watng n the queue from those beng served wthout takng n account the arrval order. Ths allows, as well as a fnte representaton, a steady state probablty whch s less detaled than the proposed n [4 whch consders the sngle staton detaled representaton wth the order of the customers n the queue, smlarly to the BCMP paper [5. On the other sde the models we propose are more detaled than those whch just consder the total number of customers n a center as the compact models of [4. The FCFS and the LCFSPR (or PS) schedulng dscplnes have dfferent GSPN representatons and the FCFS can not be used to represent the other ones f the servce depends on the customer class. The GSPN models smulate the correspondng queung system even f ther semantc s dfferent. The man dea of the defnton of the GSPN models s the way we represent the customer of the queung system whch gets the free server and the customer whch looses a server n case of preempton. In both cases we model the customer choce of a class wth a random selecton, accordng to the probablty proportonal to the number of customers n queue (or beng served) of that class over the total number of customers n queue (number of servers). The M M property allows us to state that a combnaton of GSPN-, GSPN- 2 and GSPN-3 models smlar to the servce centers combnaton n BCMP networks, has a smple closed form steady state probablty. In [ authors defne a queung center somorphc to GSPN- and show how t can be embedded n a BCMP queung network so that the steady state probablty functon of the network does not change. In the GSPN formalsm probablstc routng can be easly smulated by ntroducng a block wth a place and an mmedate transton for each possble route just after the tmed transtons of the models. Further research deals wth the extenson of the proposed LCFSPR model to Coxan servce tme dstrbutons and the defnton of algorthms to dentfy GSPN whch are compostons of models GSPN-, GSPN-2, GSPN-3 and n order to obtan effcently a set of sgnfcant performace ndexes. A Appendx Hereafter, for the sake of smplcty, by referrng to the relatons between the queueng system and GSPN models, we refer to and mx the notaton of queung theory and Petr nets. For example, we can say watng customers to refer to tokens n place P,.., P R of GSPN- or GSPN-2, customers beng served to refer to the tokens n P R+,..., P 2R of GSPN- or GSPN-2, customer arrval to refer 7
19 A. APPENDIX to token arrval, free servers to refer to the presence of tokens n the place P 2R+ or P 3R+ n the GSPN- or GSPN-2 respectvely. In our vew ths descrpton should make the followng proofs easer to understand. A. Proof of Lemma Proof. We consder four cases. Case ) Suppose that all servers are busy and at least one customer s watng, so m 2R+ = 0 and m > 0 for some =,..., R. Consder state m = (m,..., m R, m R+,..., m 2R, 0). Clearly we have that 2R =R+ m = k that s the number of servers. Consder the tangble markngs from whch m s reachable possbly through the frng of mmedate transtons combned wth tmed ones and the correspondng transton rates:: M α = {m = m e m > 0} wth rate λ M β = {m = m + e m R+ > 0} wth rate (m R+ )µp (m ); M γ = {m = m + e e R+ + e R+j m R+ > 0, j } wth rate (m R+j + )µp (m ). The set of markngs reachable from m and the correspondng transton rates can be classfed as follows: M a = {m = m + e } wth rate λ ; M b = {m = m e m R+ > 0, m > 0} wth rate (m R+ )µp (m); M c = {m = m e + e R+ e R+j m R+j > 0, m > 0, j} wth rate (m R+j )µp (m). We prove that equaton (9) satsfes the global balance equatons (GBE) by checkng that the effectve arrval rate to a state due to a new customer s equal to the effectve leavng rate due to a job completon, and the effectve arrval rate due to a job completon s equal to the effectve leavng due to new customer arrval: π(m )ξ(m, m) = π(m) ξ(m, m ) m M α m M b M c π(m )ξ(m, m) = π(m) ξ(m, m ) = m M β M γ m M a π(m )ξ(m, m) = π(m) ξ(m, m ) m M α M β M γ m M a M b M c where ξ(a, b) represents the transton rate from markng a to markng b. The effectve leavng rate from state m due to completon of a job s: [ π(m) m j+k >0 (m j+k )µ = π(m)(kµ). 8
20 A. APPENDIX The effectve arrval rate to state m due to arrvals to the system s: [ π(m )ξ(m, m) = π(m) ( m λ m M α X [ m = π(m)(kµ) R m = π(m)(kµ), j X 2R R m µ( j m j )λ ) where X = { m > 0, R}. Note that by hypothess 2R m k thus: 2R µ( m ) = kµ. Consder state m and set Y = {j m R+j > 0, j R}. The effectve leavng rate from state m due to arrvals of customers to the system s π(m) R λ. The effectve arrval rate to state m s: [ π(m) + j Y R Y j ( R λ m ) + j m j + λ j m R+ m R+j + m + ( R g= m g) + [ = π(m) j Y [ = π(m) j Y [ R = π(m) λ. µ(( 2R m ) + ) m m j + R+jµ R m + ( R g= m g) + m + µ(( 2R g= m g) + ) (m R+j + )µ λ j kµ m R+jµ + λ j m R+j k + R R Y j Y j λ j m R+ kµ µ m R+ λ j k Case 2) Consder now the case when all servers are busy, but places P,..., P R are empty. Consder a generc state m = (0,..., 0, m R+,..., m 2R, 0). The markngs from whch m s reachable are classfed as: M α = {m = m e R+ + e 2R+ m R+ > 0} wth rate λ ; M β = {m = m + e m R+ > 0} wth rate (m R+ )µp (m ); M γ = {m = m + e e R+ + e R+j m R+ > 0, j } wth rate (m R+j + )µp (m ). The markngs reachable from m and ther effectve rates are the same as n case ). Let Y be the set defned n case ). We now prove that the effectve 9
21 A. APPENDIX arrval rate to state m from markngs n M α s equal to the effectve leavng rate from m due to job completon whch s π(m)(kµ). [ π(m) ( m [ R+ µ(k)λ ) = π(m) m R+ µ = π(m)(kµ). λ k Y We prove now that the effectve arrval rate from the markngs n M β M γ s equal to the effectve leavng rate from m. [ π(m) (λ j µ(k + ) m R+jµ) + j Y [ =π(m) j Y λ j m R+j k + R Y j R Y j λ j Y m R+ λ j = π(m) k m R+ m R+j + µ(k + ) (m R+j + )µ [ R λ j. Case 3) Assume that there are not tokens n places P,..., P R and at least one server s free and at least one s busy, that s 0 < m 2R+ < k. Then the effectve leavng rate from m s smply π(m)[µ R m R+j + R λ j. The states from whch m s reachable and the correspondng rates are: M α = {m = m e R+ + e 2R+ m R+ > 0} wth rate λ ; M β = {m = m + e R+ e 2R+ } wth rate (m R+ + )µ. Let Y be defned as n case ). We now prove that the effectve arrval rate to state m s equal to the effectve leavng rate from m. [ π(m) + Y R Y λ [ =π(m) + [ =π(m) ( R m R+ R m R+j µ( λ R m R+j + m R+ + λ m R+ R m R+j R m R+j )λ µ( R m R+j + ) (m R+ + )µ R ( m R+j )µλ λ R m R+j + m R+ + R R m R+j )µ + λ. ( R m R+j + )µ (m R+ + )µ Case 4) Assume that the system s empty, that s m 2R+ = k. Ths case s trval. The effectve leavng rate from m s clearly π(m)[ R λ. The effectve arrval rate to m can just be due to a job completon and t s easy to show that s s equal to the leavng rate. 20
22 A. APPENDIX A.2 Proof of Lemma 2 Proof. The proof verfes that equaton (2) satsfes the set of GBE for the CTMC assocated to the net. We consder two cases. Case) Let m = (m,..., m R, 0,..., 0) + e R+ be a reachable tangble markng. State m can be reached from states m = m + e r+j e r+ + e due to a job completon wth rate x( + 2R r= m r)µp (m ). So the total effectve arrval rate due to a job completon s: [ R π(m) λ j + R r= m r m + x( + 2R r= m r)µ x( + 2R r= m + m r )µ + 2R r= m r R = π(m) λ j, whch s the total leavng rate from m due to arrvals to the system. State m can be reached from states m = m e j where j X = {j m j > 0, j R} wth rate λ j. So the effectve arrval rate due to an arrval to the system s: [ π(m) λ j j X m j 2R R k= m x( k k= [ 2R m k )µλ j = π(m) x( k= m k )µ, whch s dentcal to the total leavng rate due to a job completon. Case 2) If m 2R+ = the proof s trval. A.3 Proof of Lemma 4 Proof. The proof s smlar to the proof of Lemma. We prove that equaton (5) satsfes the GBE of the CTMC assocated to GSPN-2 by verfyng the partal balance. We consder agan four cases. Case ) Take a reachable markng m wth m 3R+ = 0 (.e. all servers are busy), and there s at least one token n one of the places P, wth R (.e. there s at least one customer n queue). Let X = { m > 0, R} and Y = { m R+ > 0, R}. We verfy that the effectve arrval rate to state m due to a job completon s equal to the effectve leavng rate from m due to a customer arrval and that the effectve arrval rate to state m due to a customer arrval s equal to the effectve leavng rate due to a job completon. State m s reachable due to an arrval event from states m = m e wth X Y wth rate λ p (m ) or from states m = m e + e R+ e R+j wth 2
23 A. APPENDIX X, j Y and j wth rate λ j p j (m ) Hence we can wrte: [ m π(m) R s= m µ kλ m R+ k s X Y λ + j Y [ =π(m) R s= m s [ =π(m) R s= m s X j λ j j X Y m m R+j R s= m s m R+ + µ m R+ + jkλ j k m j m R+j µ j + R s= m s µ j m R+j m j Y R [ R µ j m R+j m = π(m) µ j m R+j, X whch s dentcal to the effectve leavng rate from m due to a job completon. State m s reachable, due to job completon, from states m = m + e wth rate m R+ µ p (m ) and from states m = m + e e R+ + e R+j wth rate m R+j µ j p (m ). Hence the effectve arrval rate due to a job completon can be wrtten as: [ ( R π(m) λ m j) + m + m R+ µ m + kµ R m j + + Y X j R ( R k= λ m k) + m R+ m + j (m R+j + )µ j m + m R+j + kµ j ( R k= m k) + Y Y j [ m R+ =π(m) λ k + R Y j m R+ λ j = π(m) k [ R λ, whch s dentcal to the effectve leavng rate form state m due to a customer arrval to the system. Case 2) Consder now state m wth m 3R+ = 0 and m = 0 wth R (.e. no customers n queue). The leavng rates from m for a job completon or an arrval are the same as case. State m s reachable, due to a job completon, form states m = m + e e R+ + e R+j wth Y and j wth rate (m R+j + )µ j p j (m ) and from states m = m + e wth Y and rate m R+ µ p (m ). Hence the effectve arrval rate to m due to a job completon can be calculate as follows: [ π(m) λ m R+ µ + kµ Y [ =π(m) λ m R+ + k k Y R Y j m R+ (m R+j + )µ j = m R+j + kµ j R [ R m R+ λ j = π(m) λ j, Y j 22
24 A. APPENDIX whch s dentcal to the effectve leavng rate from m due to an arrval to the system. State m s reachable, due to an arrval to the system, from states m = m e + e 3R+ wth Y wth rate λ. Hence the effectve arrval rate to state m due to a customer arrval can be calculated as follows: [ π(m) Y m [ R+ [ R R s= m µ kλ = π(m) m R+ µ = π(m) m R+ µ, R+s λ whch s dentcal to the effectve leavng rate from m due to a job completon. Case 3) Consder the case of m wth m 3R+ = b < k. State m s reachable, due to an arrval, from states m = m e R+ + e 3R+ wth rate Y wth rate λ, thus the effectve arrval rate due to a customer arrval s: [ π(m) Y λ Y m [ R+ R e s= m µ (k b)λ = π(m) m R+ µ, R+s whch s dentcal to the effectve leavng rate from m due to a job completon. State m s reachable, due to a job completon, from states m = m+e R+ e 3R+ wth rate (m R+ + )µ, thus the effectve arrval rate due to a job completon s: [ R π(m) λ ( R s= m R+s m R+ + (m R+ + )µ µ k b + = π(m) R λ, whch s dentcal to the effectve leavng rate from m due to an arrval to the system. Case 4) consders m 3R+ = k,.e. when the system s empty, and t s trval. A.4 Proof of Lemma 5 Proof. The proof verfes that equaton (7) satsfes the set of GBE for the CTMC assocated to the net. We consder two cases. Case ) Take the tangble reachable markng m = (m,..., m R, 0,..., 0) + e R+. State m can be reached from states m = m + e r+j e r+ + e due to a job completon wth rate y j (m j + m R+j )µ j p (m ). So the effectve arrval rate to m due to a job completon s: [ R R k= π(m) λ m k + j m + m + j j (m j + )µ j y j (m j + )µ R j k= m k + = π(m) R λ j, whch s dentcal to the total leavng rate from m due to arrvals to the system. State m can be reached from state m = m + e r+j e j e R+k where j X = 23
25 A. APPENDIX {j m j > 0, j R} wth rate λ j. So the effectve arrval rate to m due to an arrval to the system s: [ m j π(m) λ R k= m y (m + m R+ )µ λ = π(m) [y (m + m R+ )µ, k j X whch s dentcal to the total leavng rate from m due to a job completon. Case 2) If m 3R+ = the proof s trval. A.5 Proof of Independence of the departure processes from GSPN models GSPN-: Proof. Consder model GSPN-. Let Θ = {(m, m) : m = m + } for =,..., R. In order to verfy equaton (20) consder a generc tangble marknf m. We consder the followng cases: ) m 2R+ = 0, and 2) m 2R+ > 0. Case ) Let m be a reachable tangble state wth m 2R+ = 0, then: Θ (, m) = {m m = m + e R+ + e j e R+j, m R+j > 0, j } {m m = m + e, m R+ > 0},, j R. Let sgn(m ) be the ndcator functon defned as follows: sgn(m ) = when m > 0 and 0 otherwse, and let Y = {j m R+j > 0, j R}. The left hand sde of equaton (20) can be rewrtten as follows: [ π(m) j Y j λ + R a= m a m j + m R+j m R+ + kµ (m R+ + )µ + R a= + sgn(m )λ m a m + kµ m m + R+µ + R a= m a ( ) = π(m) [λ = π(m)λ, j Y j m R+j k + m R+ k m j + + R a= m a whch gves the rght hand sde of equaton (20). Case 2) Le m be a reachable tangble state wth m 2R+ > 0. Then Θ (, m) = {m + e R+ e 2R+ }. Thus equaton (20) holds, n fact: + π(m) [λ R a= m R+a m R+ + ( + R a= m R+a)µ ( + m R+)µ = π(m)λ. Ths proves that the traffc processes assocated to Θ, R, are pontwse ndependent Posson processes,.e., the departure processes for each class of customers are Posson ndependent processes under ndependent Posson arrvals. 24
26 A. APPENDIX GSPN-2: Proof. Consder model GSPN-2. In ths model m = m +m R+, wth R and let Θ = {(m, m) : m = m +}. In order to verfy equaton (20) consde a generc tangble state m. We consder the followng cases: ) m 3R+ = 0, and 2) m 3R+ > 0. Case ) Let m be a reachable tangble state wth m 3R+ = 0, then: Θ (, m) = {m m = m + e R+ + e j e R+j, m R+j > 0, j } {m m = m + e m R+ > 0}, j R. Let sgn(m ) be the ndcator functon defned as follows: sgn(m ) = when m > 0 and sgn(m ) = 0 otherwse, and let Y = {j m R+j > 0, j R}. The left hand sde of equaton (20) can be rewrtten as follows: [ π(m) j Y j λ + R a= m a m j + + R a= + sgn(m )λ m a m + ( =π(m) [λ j Y j m R+j k + m R+ k m R+j m j + (m R+ + )µ m R+ + kµ + R m + m R+ µ kµ + R ) = π(m)λ, a= m a a= m a whch s the rght hand sde of equaton (20). Case 2) Le m be a reachable tangble state wth m 3R+ > 0. Then Θ (, m) = {m + e R+ e 3R+ }. Thus equaton (20) holds, n fact: + R a= π(m)[λ m a (m R+ + )µ = π(m)λ. m R+ + (m R+ + )µ GSPN-3 Proof. Consder model GSPN-3. Let Θ = {(m, m) m = m + e, R}. Provng that equaton (20) holds s trval. References. Afshar, P. V., Bruell, S. C., and Kan, R. Y. Modelng a new technque for accessng shared buses. In Proceedngs of the Computer Network Performance Symposum (New York, NY, USA, 982), ACM Press, pp Balbo, G., Bruell, S. C., and Ghanta, S. Combnng queueng network and generalzed stochastc Petr nets for the soluton of complex models of system behavor. IEEE Transactons on Computers 37 (998),
27 A. APPENDIX 3. Balbo, G., Bruell, S. C., and Sereno, M. Product form soluton for Generalzed Stochastc Petr Nets. IEEE Transactons on Software Engneerng 28 (2002), Balbo, G., Bruell, S. C., and Sereno, M. On the relatons between BCMP Queueng Networks and Product Form Soluton Stochastc Petr Nets. 0th Internatonal Workshop on Petr Nets and Performance Models, Proceedngs (2003), Baskett, F., Chandy, K. M., Muntz, R. R., and Palacos, F. G. Open, closed, and mxed networks of queues wth dfferent classes of customers. J. ACM 22, 2 (975), Bruell, S. C., G.Balbo, and Afshar, P. V. Mean Value Analyss of mxed, multple class BCMP networks wth load dependent servce statons. Performance Evaluaton 4 (984), Buzen, J. P. Computatonal algorthms for closed queueng networks wth exponental servers. Commun. ACM 6, 9 (973), Chandy, K. M., and Sauer, C. H. Computatonal algorthms for product form queueng networks. Commun. ACM 23, 0 (980), Chola, G., Marsan, M. A., Balbo, G., and Conte, G. Generalzed stochastc Petr nets: a defnton at the net level and ts mplcatons. IEEE Transactons on software engneerng 9, 2 (993), Cohen, J. W. The sngle server queue. Wley-Interscence, Coleman, J. L., Henderson, W., and Taylor, P. G. Product form equlbrum dstrbutons and a convoluton algorthm for Stochastc Petr nets. Performance Evaluaton 26 (996), Henderson, W., Lucc, D., and Taylor, P. G. A net level performance analyss of Stochastc Petr Nets. J. Austral. Math. Soc. Ser. B 3 (989), Kant, K. Introducton to Computer System Performance Evaluaton. McGraw- Hll, 992, ch. 4 and Klenrock, L. Queueng Systems, vol. (Theory). John Wley and Sons, Marsan, M. A., Balbo, G., Conte, G., Donatell, S., and Franceschns, G. Modellng wth generalzed stochastc Petr nets. Wley, Melamed, B. On Posson traffc processes n dscrete-state markovan systems wth applcatons to queueng theory. Advances n Appled Probablty, (979), Muntz, R. R. Posson departure processes and queueng networks. Tech. Rep. IBM Research Report RC445, Yorktown Heghts, New York, Resser, M., and Lavenberg, S. S. Mean Value Analyss of closed multchan queueng network. JACM 27, 2 (980), Taylor, H. M., and Karln, S. An Introducton To Stochastc Modelng, 3rd ed. Academc Press, 998, ch. IX. 20. Vernon, M., Zahorjan, J., and Lazowska, E. D. A comparson of performance Petr Nets and queueng network models. Proceedngs 3rd Intern. Workshop on Modellng Technques and Performance Evaluaton (987),
MODELING TRAFFIC LIGHTS IN INTERSECTION USING PETRI NETS
The 3 rd Internatonal Conference on Mathematcs and Statstcs (ICoMS-3) Insttut Pertanan Bogor, Indonesa, 5-6 August 28 MODELING TRAFFIC LIGHTS IN INTERSECTION USING PETRI NETS 1 Deky Adzkya and 2 Subono
More informationAnalysis of Discrete Time Queues (Section 4.6)
Analyss of Dscrete Tme Queues (Secton 4.6) Copyrght 2002, Sanjay K. Bose Tme axs dvded nto slots slot slot boundares Arrvals can only occur at slot boundares Servce to a job can only start at a slot boundary
More informationQueueing Networks II Network Performance
Queueng Networks II Network Performance Davd Tpper Assocate Professor Graduate Telecommuncatons and Networkng Program Unversty of Pttsburgh Sldes 6 Networks of Queues Many communcaton systems must be modeled
More informationIntroduction to Continuous-Time Markov Chains and Queueing Theory
Introducton to Contnuous-Tme Markov Chans and Queueng Theory From DTMC to CTMC p p 1 p 12 1 2 k-1 k p k-1,k p k-1,k k+1 p 1 p 21 p k,k-1 p k,k-1 DTMC 1. Transtons at dscrete tme steps n=,1,2, 2. Past doesn
More informationTCOM 501: Networking Theory & Fundamentals. Lecture 7 February 25, 2003 Prof. Yannis A. Korilis
TCOM 501: Networkng Theory & Fundamentals Lecture 7 February 25, 2003 Prof. Yanns A. Korls 1 7-2 Topcs Open Jackson Networks Network Flows State-Dependent Servce Rates Networks of Transmsson Lnes Klenrock
More informationEquilibrium Analysis of the M/G/1 Queue
Eulbrum nalyss of the M/G/ Queue Copyrght, Sanay K. ose. Mean nalyss usng Resdual Lfe rguments Secton 3.. nalyss usng an Imbedded Marov Chan pproach Secton 3. 3. Method of Supplementary Varables done later!
More informationDifference Equations
Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1
More informationStructure and Drive Paul A. Jensen Copyright July 20, 2003
Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.
More informationLectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix
Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could
More informationMarkov chains. Definition of a CTMC: [2, page 381] is a continuous time, discrete value random process such that for an infinitesimal
Markov chans M. Veeraraghavan; March 17, 2004 [Tp: Study the MC, QT, and Lttle s law lectures together: CTMC (MC lecture), M/M/1 queue (QT lecture), Lttle s law lecture (when dervng the mean response tme
More informationMeenu Gupta, Man Singh & Deepak Gupta
IJS, Vol., o. 3-4, (July-December 0, pp. 489-497 Serals Publcatons ISS: 097-754X THE STEADY-STATE SOLUTIOS OF ULTIPLE PARALLEL CHAELS I SERIES AD O-SERIAL ULTIPLE PARALLEL CHAELS BOTH WITH BALKIG & REEGIG
More informationA new Approach for Solving Linear Ordinary Differential Equations
, ISSN 974-57X (Onlne), ISSN 974-5718 (Prnt), Vol. ; Issue No. 1; Year 14, Copyrght 13-14 by CESER PUBLICATIONS A new Approach for Solvng Lnear Ordnary Dfferental Equatons Fawz Abdelwahd Department of
More informationCOMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
More informationLecture 4: November 17, Part 1 Single Buffer Management
Lecturer: Ad Rosén Algorthms for the anagement of Networs Fall 2003-2004 Lecture 4: November 7, 2003 Scrbe: Guy Grebla Part Sngle Buffer anagement In the prevous lecture we taled about the Combned Input
More informationNotes on Frequency Estimation in Data Streams
Notes on Frequency Estmaton n Data Streams In (one of) the data streamng model(s), the data s a sequence of arrvals a 1, a 2,..., a m of the form a j = (, v) where s the dentty of the tem and belongs to
More informationAdditional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty
Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,
More informationInner Product. Euclidean Space. Orthonormal Basis. Orthogonal
Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,
More informationEcon107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)
I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes
More informationFoundations of Arithmetic
Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More informationECE559VV Project Report
ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate
More informationHidden Markov Models & The Multivariate Gaussian (10/26/04)
CS281A/Stat241A: Statstcal Learnng Theory Hdden Markov Models & The Multvarate Gaussan (10/26/04) Lecturer: Mchael I. Jordan Scrbes: Jonathan W. Hu 1 Hdden Markov Models As a bref revew, hdden Markov models
More informationCHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE
CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng
More informationApplied Stochastic Processes
STAT455/855 Fall 23 Appled Stochastc Processes Fnal Exam, Bref Solutons 1. (15 marks) (a) (7 marks) The dstrbuton of Y s gven by ( ) ( ) y 2 1 5 P (Y y) for y 2, 3,... The above follows because each of
More information6. Stochastic processes (2)
6. Stochastc processes () Lect6.ppt S-38.45 - Introducton to Teletraffc Theory Sprng 5 6. Stochastc processes () Contents Markov processes Brth-death processes 6. Stochastc processes () Markov process
More information6. Stochastic processes (2)
Contents Markov processes Brth-death processes Lect6.ppt S-38.45 - Introducton to Teletraffc Theory Sprng 5 Markov process Consder a contnuous-tme and dscrete-state stochastc process X(t) wth state space
More information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More informationLinear Approximation with Regularization and Moving Least Squares
Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...
More informationCanonical transformations
Canoncal transformatons November 23, 2014 Recall that we have defned a symplectc transformaton to be any lnear transformaton M A B leavng the symplectc form nvarant, Ω AB M A CM B DΩ CD Coordnate transformatons,
More informationFor now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.
Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson
More informationThe Feynman path integral
The Feynman path ntegral Aprl 3, 205 Hesenberg and Schrödnger pctures The Schrödnger wave functon places the tme dependence of a physcal system n the state, ψ, t, where the state s a vector n Hlbert space
More informationMore metrics on cartesian products
More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of
More informationPhysics 5153 Classical Mechanics. D Alembert s Principle and The Lagrangian-1
P. Guterrez Physcs 5153 Classcal Mechancs D Alembert s Prncple and The Lagrangan 1 Introducton The prncple of vrtual work provdes a method of solvng problems of statc equlbrum wthout havng to consder the
More informationConvergence of random processes
DS-GA 12 Lecture notes 6 Fall 216 Convergence of random processes 1 Introducton In these notes we study convergence of dscrete random processes. Ths allows to characterze phenomena such as the law of large
More informationCHAPTER III Neural Networks as Associative Memory
CHAPTER III Neural Networs as Assocatve Memory Introducton One of the prmary functons of the bran s assocatve memory. We assocate the faces wth names, letters wth sounds, or we can recognze the people
More informationFinding Primitive Roots Pseudo-Deterministically
Electronc Colloquum on Computatonal Complexty, Report No 207 (205) Fndng Prmtve Roots Pseudo-Determnstcally Ofer Grossman December 22, 205 Abstract Pseudo-determnstc algorthms are randomzed search algorthms
More informationContinuous Time Markov Chains
Contnuous Tme Markov Chans Brth and Death Processes,Transton Probablty Functon, Kolmogorov Equatons, Lmtng Probabltes, Unformzaton Chapter 6 1 Markovan Processes State Space Parameter Space (Tme) Dscrete
More informationModule 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur
Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:
More informationLecture Space-Bounded Derandomization
Notes on Complexty Theory Last updated: October, 2008 Jonathan Katz Lecture Space-Bounded Derandomzaton 1 Space-Bounded Derandomzaton We now dscuss derandomzaton of space-bounded algorthms. Here non-trval
More informationLecture Notes on Linear Regression
Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume
More informationVapnik-Chervonenkis theory
Vapnk-Chervonenks theory Rs Kondor June 13, 2008 For the purposes of ths lecture, we restrct ourselves to the bnary supervsed batch learnng settng. We assume that we have an nput space X, and an unknown
More informationComputing Correlated Equilibria in Multi-Player Games
Computng Correlated Equlbra n Mult-Player Games Chrstos H. Papadmtrou Presented by Zhanxang Huang December 7th, 2005 1 The Author Dr. Chrstos H. Papadmtrou CS professor at UC Berkley (taught at Harvard,
More informationApplication of Queuing Theory to Waiting Time of Out-Patients in Hospitals.
Applcaton of Queung Theory to Watng Tme of Out-Patents n Hosptals. R.A. Adeleke *, O.D. Ogunwale, and O.Y. Hald. Department of Mathematcal Scences, Unversty of Ado-Ekt, Ado-Ekt, Ekt State, Ngera. E-mal:
More informationContinuous Time Markov Chain
Contnuous Tme Markov Chan Hu Jn Department of Electroncs and Communcaton Engneerng Hanyang Unversty ERICA Campus Contents Contnuous tme Markov Chan (CTMC) Propertes of sojourn tme Relatons Transton probablty
More informationLecture 2: Gram-Schmidt Vectors and the LLL Algorithm
NYU, Fall 2016 Lattces Mn Course Lecture 2: Gram-Schmdt Vectors and the LLL Algorthm Lecturer: Noah Stephens-Davdowtz 2.1 The Shortest Vector Problem In our last lecture, we consdered short solutons to
More informationPower law and dimension of the maximum value for belief distribution with the max Deng entropy
Power law and dmenson of the maxmum value for belef dstrbuton wth the max Deng entropy Bngy Kang a, a College of Informaton Engneerng, Northwest A&F Unversty, Yanglng, Shaanx, 712100, Chna. Abstract Deng
More informationComparison of the Population Variance Estimators. of 2-Parameter Exponential Distribution Based on. Multiple Criteria Decision Making Method
Appled Mathematcal Scences, Vol. 7, 0, no. 47, 07-0 HIARI Ltd, www.m-hkar.com Comparson of the Populaton Varance Estmators of -Parameter Exponental Dstrbuton Based on Multple Crtera Decson Makng Method
More information1 GSW Iterative Techniques for y = Ax
1 for y = A I m gong to cheat here. here are a lot of teratve technques that can be used to solve the general case of a set of smultaneous equatons (wrtten n the matr form as y = A), but ths chapter sn
More informationCHAPTER 17 Amortized Analysis
CHAPTER 7 Amortzed Analyss In an amortzed analyss, the tme requred to perform a sequence of data structure operatons s averaged over all the operatons performed. It can be used to show that the average
More informationPHYS 705: Classical Mechanics. Canonical Transformation II
1 PHYS 705: Classcal Mechancs Canoncal Transformaton II Example: Harmonc Oscllator f ( x) x m 0 x U( x) x mx x LT U m Defne or L p p mx x x m mx x H px L px p m p x m m H p 1 x m p m 1 m H x p m x m m
More informationSystem in Weibull Distribution
Internatonal Matheatcal Foru 4 9 no. 9 94-95 Relablty Equvalence Factors of a Seres-Parallel Syste n Webull Dstrbuton M. A. El-Dacese Matheatcs Departent Faculty of Scence Tanta Unversty Tanta Egypt eldacese@yahoo.co
More informationThe Geometry of Logit and Probit
The Geometry of Logt and Probt Ths short note s meant as a supplement to Chapters and 3 of Spatal Models of Parlamentary Votng and the notaton and reference to fgures n the text below s to those two chapters.
More informationNUMERICAL DIFFERENTIATION
NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the
More informationOnline Appendix. t=1 (p t w)q t. Then the first order condition shows that
Artcle forthcomng to ; manuscrpt no (Please, provde the manuscrpt number!) 1 Onlne Appendx Appendx E: Proofs Proof of Proposton 1 Frst we derve the equlbrum when the manufacturer does not vertcally ntegrate
More informationNumerical Heat and Mass Transfer
Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and
More informationThe Order Relation and Trace Inequalities for. Hermitian Operators
Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence
More informationA new construction of 3-separable matrices via an improved decoding of Macula s construction
Dscrete Optmzaton 5 008 700 704 Contents lsts avalable at ScenceDrect Dscrete Optmzaton journal homepage: wwwelsevercom/locate/dsopt A new constructon of 3-separable matrces va an mproved decodng of Macula
More informationSimulated Power of the Discrete Cramér-von Mises Goodness-of-Fit Tests
Smulated of the Cramér-von Mses Goodness-of-Ft Tests Steele, M., Chaselng, J. and 3 Hurst, C. School of Mathematcal and Physcal Scences, James Cook Unversty, Australan School of Envronmental Studes, Grffth
More informationGames of Threats. Elon Kohlberg Abraham Neyman. Working Paper
Games of Threats Elon Kohlberg Abraham Neyman Workng Paper 18-023 Games of Threats Elon Kohlberg Harvard Busness School Abraham Neyman The Hebrew Unversty of Jerusalem Workng Paper 18-023 Copyrght 2017
More informationRandom Walks on Digraphs
Random Walks on Dgraphs J. J. P. Veerman October 23, 27 Introducton Let V = {, n} be a vertex set and S a non-negatve row-stochastc matrx (.e. rows sum to ). V and S defne a dgraph G = G(V, S) and a drected
More informationInformation Geometry of Gibbs Sampler
Informaton Geometry of Gbbs Sampler Kazuya Takabatake Neuroscence Research Insttute AIST Central 2, Umezono 1-1-1, Tsukuba JAPAN 305-8568 k.takabatake@ast.go.jp Abstract: - Ths paper shows some nformaton
More information1 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations
Physcs 171/271 -Davd Klenfeld - Fall 2005 (revsed Wnter 2011) 1 Dervaton of Rate Equatons from Sngle-Cell Conductance (Hodgkn-Huxley-lke) Equatons We consder a network of many neurons, each of whch obeys
More informationAPPENDIX A Some Linear Algebra
APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,
More informationNON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS
IJRRAS 8 (3 September 011 www.arpapress.com/volumes/vol8issue3/ijrras_8_3_08.pdf NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS H.O. Bakodah Dept. of Mathematc
More informationCase A. P k = Ni ( 2L i k 1 ) + (# big cells) 10d 2 P k.
THE CELLULAR METHOD In ths lecture, we ntroduce the cellular method as an approach to ncdence geometry theorems lke the Szemeréd-Trotter theorem. The method was ntroduced n the paper Combnatoral complexty
More informationDETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM
Ganj, Z. Z., et al.: Determnaton of Temperature Dstrbuton for S111 DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM by Davood Domr GANJI
More informationPERFORMANCE OF MULTICLASS MARKOVIAN QUEUEING NETWORKS VIA PIECEWISE LINEAR LYAPUNOV FUNCTIONS 1
The Annals of Appled Probablty 2001, Vol. 11, No. 4, 1384 1428 PERFORMANCE OF MULTICLASS MARKOVIAN QUEUEING NETWORKS VIA PIECEWISE LINEAR LYAPUNOV FUNCTIONS 1 By Dmtrs Bertsmas, Davd Gamarnk and John N.
More informationA comment on Boucherie product-form results
A comment on Boucherie product-form results Andrea Marin Dipartimento di Informatica Università Ca Foscari di Venezia Via Torino 155, 30172 Venezia Mestre, Italy {balsamo,marin}@dsi.unive.it Abstract.
More informationRandomness and Computation
Randomness and Computaton or, Randomzed Algorthms Mary Cryan School of Informatcs Unversty of Ednburgh RC 208/9) Lecture 0 slde Balls n Bns m balls, n bns, and balls thrown unformly at random nto bns usually
More informationTHE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens
THE CHINESE REMAINDER THEOREM KEITH CONRAD We should thank the Chnese for ther wonderful remander theorem. Glenn Stevens 1. Introducton The Chnese remander theorem says we can unquely solve any par of
More informationCHAPTER 4. Vector Spaces
man 2007/2/16 page 234 CHAPTER 4 Vector Spaces To crtcze mathematcs for ts abstracton s to mss the pont entrel. Abstracton s what makes mathematcs work. Ian Stewart The man am of ths tet s to stud lnear
More informationGeneralized Linear Methods
Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set
More informationMean value analysis of product form solution queueing networks with repetitive service blocking
Performance Evaluaton 36 37 (1999) 19 33 www.elsever.com/locate/peva Mean value analyss of product form soluton queueng networks wth repettve servce blockng Matteo Sereno Dpartmento d Informatca, Unverstà
More informationA Hybrid Variational Iteration Method for Blasius Equation
Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 1932-9466 Vol. 10, Issue 1 (June 2015), pp. 223-229 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) A Hybrd Varatonal Iteraton Method
More informationKernel Methods and SVMs Extension
Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general
More informationTransfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system
Transfer Functons Convenent representaton of a lnear, dynamc model. A transfer functon (TF) relates one nput and one output: x t X s y t system Y s The followng termnology s used: x y nput output forcng
More information1 Matrix representations of canonical matrices
1 Matrx representatons of canoncal matrces 2-d rotaton around the orgn: ( ) cos θ sn θ R 0 = sn θ cos θ 3-d rotaton around the x-axs: R x = 1 0 0 0 cos θ sn θ 0 sn θ cos θ 3-d rotaton around the y-axs:
More informationChapter - 2. Distribution System Power Flow Analysis
Chapter - 2 Dstrbuton System Power Flow Analyss CHAPTER - 2 Radal Dstrbuton System Load Flow 2.1 Introducton Load flow s an mportant tool [66] for analyzng electrcal power system network performance. Load
More informationSupplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso
Supplement: Proofs and Techncal Detals for The Soluton Path of the Generalzed Lasso Ryan J. Tbshran Jonathan Taylor In ths document we gve supplementary detals to the paper The Soluton Path of the Generalzed
More informationIntroduction to Vapor/Liquid Equilibrium, part 2. Raoult s Law:
CE304, Sprng 2004 Lecture 4 Introducton to Vapor/Lqud Equlbrum, part 2 Raoult s Law: The smplest model that allows us do VLE calculatons s obtaned when we assume that the vapor phase s an deal gas, and
More informationEEE 241: Linear Systems
EEE : Lnear Systems Summary #: Backpropagaton BACKPROPAGATION The perceptron rule as well as the Wdrow Hoff learnng were desgned to tran sngle layer networks. They suffer from the same dsadvantage: they
More informationMin Cut, Fast Cut, Polynomial Identities
Randomzed Algorthms, Summer 016 Mn Cut, Fast Cut, Polynomal Identtes Instructor: Thomas Kesselhem and Kurt Mehlhorn 1 Mn Cuts n Graphs Lecture (5 pages) Throughout ths secton, G = (V, E) s a mult-graph.
More informationSection 8.3 Polar Form of Complex Numbers
80 Chapter 8 Secton 8 Polar Form of Complex Numbers From prevous classes, you may have encountered magnary numbers the square roots of negatve numbers and, more generally, complex numbers whch are the
More information2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification
E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton
More informationLecture 10 Support Vector Machines II
Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed
More informationj) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1
Random varables Measure of central tendences and varablty (means and varances) Jont densty functons and ndependence Measures of assocaton (covarance and correlaton) Interestng result Condtonal dstrbutons
More informationBounds on the bias terms for the Markov reward approach
Bounds on the bas terms for the Markov reward approach Xnwe Ba 1 and Jasper Goselng 1 arxv:1901.00677v1 [math.pr] 3 Jan 2019 1 Department of Appled Mathematcs, Unversty of Twente, P.O. Box 217, 7500 AE
More information2.3 Nilpotent endomorphisms
s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms
More informationThe Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction
ECONOMICS 5* -- NOTE (Summary) ECON 5* -- NOTE The Multple Classcal Lnear Regresson Model (CLRM): Specfcaton and Assumptons. Introducton CLRM stands for the Classcal Lnear Regresson Model. The CLRM s also
More informationarxiv: v1 [math.co] 12 Sep 2014
arxv:1409.3707v1 [math.co] 12 Sep 2014 On the bnomal sums of Horadam sequence Nazmye Ylmaz and Necat Taskara Department of Mathematcs, Scence Faculty, Selcuk Unversty, 42075, Campus, Konya, Turkey March
More informationSalmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2
Salmon: Lectures on partal dfferental equatons 5. Classfcaton of second-order equatons There are general methods for classfyng hgher-order partal dfferental equatons. One s very general (applyng even to
More informationLecture 17 : Stochastic Processes II
: Stochastc Processes II 1 Contnuous-tme stochastc process So far we have studed dscrete-tme stochastc processes. We studed the concept of Makov chans and martngales, tme seres analyss, and regresson analyss
More information9 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations
Physcs 171/271 - Chapter 9R -Davd Klenfeld - Fall 2005 9 Dervaton of Rate Equatons from Sngle-Cell Conductance (Hodgkn-Huxley-lke) Equatons We consder a network of many neurons, each of whch obeys a set
More informationText S1: Detailed proofs for The time scale of evolutionary innovation
Text S: Detaled proofs for The tme scale of evolutonary nnovaton Krshnendu Chatterjee Andreas Pavloganns Ben Adlam Martn A. Nowak. Overvew and Organzaton We wll present detaled proofs of all our results.
More informationMarkov Chain Monte Carlo Lecture 6
where (x 1,..., x N ) X N, N s called the populaton sze, f(x) f (x) for at least one {1, 2,..., N}, and those dfferent from f(x) are called the tral dstrbutons n terms of mportance samplng. Dfferent ways
More informationThe optimal delay of the second test is therefore approximately 210 hours earlier than =2.
THE IEC 61508 FORMULAS 223 The optmal delay of the second test s therefore approxmately 210 hours earler than =2. 8.4 The IEC 61508 Formulas IEC 61508-6 provdes approxmaton formulas for the PF for smple
More information9 Characteristic classes
THEODORE VORONOV DIFFERENTIAL GEOMETRY. Sprng 2009 [under constructon] 9 Characterstc classes 9.1 The frst Chern class of a lne bundle Consder a complex vector bundle E B of rank p. We shall construct
More informationOpen Systems: Chemical Potential and Partial Molar Quantities Chemical Potential
Open Systems: Chemcal Potental and Partal Molar Quanttes Chemcal Potental For closed systems, we have derved the followng relatonshps: du = TdS pdv dh = TdS + Vdp da = SdT pdv dg = VdP SdT For open systems,
More informationA New Refinement of Jacobi Method for Solution of Linear System Equations AX=b
Int J Contemp Math Scences, Vol 3, 28, no 17, 819-827 A New Refnement of Jacob Method for Soluton of Lnear System Equatons AX=b F Naem Dafchah Department of Mathematcs, Faculty of Scences Unversty of Gulan,
More informationErrors for Linear Systems
Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch
More information